Skip to main content

Biren BR20X (Next-Generation Flagship Data Center GPU)

Product Overview​

BR20X is Biren Technology's next-generation flagship data center chip, with tape-out planned for Q3 2026 and official commercial launch in the second half of the year.

Compared with the current BR100 / BR104 (Bili series), the BR20X's upgrade focus is not single-point compute but the breadth of native support for low-precision data formats — with an emphasis on enhanced native support for FP8, FP4, and other wider data formats, to match the demand of large model inference for low-precision compute.

Biren is one of the last members of the Chinese GPU "Four Little Dragons" (Enflame, Moore Threads, MetaX, Biren) to complete capitalization. All four are now listed, and the BR20X is Biren's first flagship product launched after its capital arrival; its tape-out and commercialization cadence is a key window into the iteration capability of high-end Chinese GPUs.

Core Specifications​

ParameterValue
ArchitectureNot disclosed (BR20X new-generation architecture)
Process NodeNot disclosed
PositioningData center large model training / inference
Natively Supported FormatsFocused enhancement of low-precision formats such as FP8 and FP4
FP16/BF16 ComputeNot disclosed
Memory CapacityNot disclosed
Memory BandwidthNot disclosed
TDPNot disclosed
Tape-outPlanned 2026-Q3
LaunchCommercial launch planned second half of 2026 (in development)

⚠️ Specification notes: The BR20X is currently in development, and Biren has not published any compute, memory, or power specifications. The "2026-Q3 tape-out", "second-half commercialization", and "enhanced native FP8/FP4 support" entries in the table come from public reports and company disclosures.

Next-Generation Chinese GPU Product Timeline​

From the second half of 2026 through 2027 is the key window in which next-generation Chinese GPU products land in quick succession and move from tape-out validation to scaled commercial deployment:

VendorNext-Generation ProductKey MilestonesData Formats
BirenBR20X2026-Q3 tape-out, second-half commercializationNative FP8 / FP4
MetaXXiyun C7002026 second-half tape-out, 2027 second-half mass productionAdds FP4
Moore ThreadsHuashan (AI) / Lushan (graphics)Mass production / launch within 20265th-gen Huagang architecture
Enflame5th-generation cloud AI chipAnnounced in 2027FP4 support, cross-modal generation

Key insight: The four vendors' timelines are highly convergent — tape-out in the second half of 2026, mass production in 2027 is the common rhythm. This shows that the competitive focus of Chinese GPUs has shifted from "who can build it" to "who can tape out on schedule and deliver at scale". For buyers, 2027 is the year high-end Chinese GPUs become truly evaluable, while 2026 remains a transition period dominated by current mass-produced models (Ascend 910C, MLU690, MTT S5000, Xiyun C600).

Relationship to the Current BR Series​

ParameterBR100 / BR104BR20X
Architecture GenerationFirst-generation Bili architectureNew generation (undisclosed)
StatusReleasedIn development
Low-Precision SupportYes (see card for specific range)Focused enhancement of FP8 / FP4
Main ScenariosData center AI training/inferenceData center AI training/inference
Purchase AdviceEvaluableEvaluate after tape-out

Vendor Information​

ItemDetails
CompanyShanghai Biren Technology Co., Ltd.
HeadquartersShanghai
Founded2019-09
IPOSTAR Market (listed)
Core ProductsBili (BR) series general-purpose GPUs, BLink interconnect
Industry PositioningOne of the Chinese GPU "Four Little Dragons"

Use Cases​

  • ✅ Data center large model training and inference (planned positioning)
  • ✅ Chinese compute clusters (paired with the BLink 2.0 supernode solution)
  • ✅ Low-precision inference scenarios (native FP8 / FP4 is the key direction)
  • ❌ No products currently on sale (in development, specifications undecided)
  • ❌ Selection reference must await official data after tape-out

References​